Engineering Fracture Mechanics 155 (2016) 49–66

Contents lists available at ScienceDirect

Engineering Fracture Mechanics

journal homepage: www.elsevier.com/locate/engfracmech

An isotropic damage model based on fracture mechanics for

concrete
Mao Kurumatani a,⇑, Kenjiro Terada b, Junji Kato c, Takashi Kyoya c, Kazuo Kashiyama d
a
Department of Urban and Civil Engineering, Ibaraki University, 4-12-1 Nakanarusawa, Hitachi, Ibaraki 316-8511, Japan
b
International Research Institute of Disaster Science, Tohoku University, 468-1 Aza-Aoba, Aramaki, Aoba, Sendai 980-0845, Japan
c
Department of Civil Engineering, Tohoku University, 6-6-06 Aza-Aoba, Aramaki, Aoba, Sendai 980-8579, Japan
d
Department of Civil and Environmental Engineering, Chuo University, 1-13-27 Kasuga, Bunkyo-ku, Tokyo 112-8551, Japan

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents an isotropic damage model for quasi-brittle materials and demon-
Received 6 August 2015 strates its performance in crack propagation analysis for concrete. The suggested damage
Received in revised form 13 January 2016 model is based on fracture mechanics for concrete in terms of the fracture energy and is
Accepted 14 January 2016
therefore capable of simulating strain-softening behavior without mesh-size dependency.
Available online 22 January 2016
After the cohesive crack model is incorporated into a damage model in a 1D setting, the 1D
model is extended to multi-dimensional problems by means of the equivalent strain of the
Keywords:
modified von-Mises model. Several numerical examples are presented to examine the fun-
Isotropic damage model
Concrete
damental characteristics of the proposed damage model. In particular, we demonstrate
Fracture mechanics that finite element solutions with the damage model are independent of mesh size and that
Traction–separation law the energy balance evaluated for the fracture process in a three-point bending test for a
Crack propagation analysis specimen with a single-edge notch is consistent with the fracture energy. A benchmark test
for a mixed-mode fracture is also conducted to demonstrate the performance of the pro-
posed damage model.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Materials such as concrete and ground/bedrock are not completely brittle materials, as they exhibit quasi-brittle fracture
behavior, which results in failure with strain-softening. To reproduce the fracture behavior of such quasi-brittle materials
efficiently and accurately, a considerable number of studies have been made on the developments of discretization methods
and constitutive models, which still remain matter for debate.
Numerical techniques for fracture analysis with the finite element method (FEM) can roughly be classified into two
groups. One is a class of constitutive models that represent strain-softening behavior associated with fracture, such as the
smeared crack model and the damage model, whereas the other is a class of methods that incorporate discontinuous dis-
placement fields into finite element (FE) approximations. In recent years, the latter class has been studied actively in the field
of computational mechanics. It includes, for example, the embedded finite element method [1], which incorporates a discon-
tinuous incompatible mode into a shape function in the FEM, the extended finite element method (X-FEM) [2] and the finite
cover method (FCM) [3], both of which are based on the method of partition-of-unity (PU). The performances of these meth-
ods to capture arbitrary discontinuities independently of FE mesh and to evaluate strain-softening behavior have also been

⇑ Corresponding author. Tel.: +81 294 38 5162; fax: +81 294 38 5268.
E-mail address: [email protected] (M. Kurumatani).

http://dx.doi.org/10.1016/j.engfracmech.2016.01.020
0013-7944/Ó 2016 Elsevier Ltd. All rights reserved.
50 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

Nomenclature

r 1D Cauchy stress
e 1D small strain
E Young’s modulus
D scalar damage variable ð0 6 D 6 1Þ
f 1D cohesive force
w 1D crack opening displacement
A; B unknown material parameters
ft uniaxial tensile strength
Gf fracture energy
e0 damage initiation strain e0 ¼ f t =E
h element length in 1D
he representative element length in 2D or 3D
 1D maximum strain in the deformation history
a; b material parameters in other references
eeq equivalent strain defined as the modified von-Mises criterion
m Poisson’s ratio
k ratio of tensile and compressive strength
e small strain tensor
e deviatoric strain tensor
I1 first invariant of strain tensor e
J2 second invariant of deviatoric strain tensor e
r Cauchy stress tensor
c elastic modulus tensor
j maximum equivalent strain in the deformation history
Ae area of 2D elements
Ve volume of 3D elements
C material Jacobian
a1 ; a2 coefficients in Eq. (34)
GF apparent fracture energy estimated from numerical analyses

presented [4,5]. Although these methods are better suited for representing discontinuous displacements with high accuracy,
the tracing of crack geometries seems to make it difficult to apply these methods to problems in which multiple cracks occur
and interact with each other, or to 3D problems.
On the other hand, the former methods, which replace cracking and fracturing by material behavior characterized by
appropriate constitutive relations, have been studied for a long time because of their good compatibility with the FEM.
One of the representative methods for concrete is the crack band model [6], which expresses strain-softening behavior using
tensile strength, fracture energy, and damage region width as parameters. The rotating crack model [7], which expresses
crack nucleation/propagation and resulting softening behavior with the principal stress and its direction, is also one of
the most fundamental models, and is capable of reproducing the anisotropy in crack formation. It is, however, known that
these methods based on continuum mechanics often cause uncalled-for problems such as stress locking, depending on the FE
mesh or crack geometry [7,8].
In the rotating crack model, the strain tensor is decomposed into elastic and inelastic parts to represent the
strain-softening behavior. In contrast, in damage models, the strain-softening behavior is expressed by multiplying damage
variables that represent the degree of mechanical degradation by elastic constants, without introduction of inelastic strains.
Representative examples of these models include the isotropic damage model, in which damage variables are expressed as
scalar-valued functions, and the anisotropic model based on the microplane model [9]. Although these models do not
directly describe discontinuous displacements, they have proven to be effective in accurately reproducing the experimental
results of concrete crack behavior [10,11]. In the damage evolution laws, however, several parameters irrespective of mate-
rial responses have to be set to determine the functional form to represent strain-softening [12–14].
It is also known that incorporating a rotating crack model or a damage model into the FEM to analyze the strain-softening
behavior entails the dependency of analysis results on element type or element size [15]. The so-called nonlocal models such
as the integral average model [16] and the gradient model [17] are known to be common techniques to overcome this prob-
lem, namely, the lack of mesh-objectivity. Although applying these techniques enable us to reduce mesh dependency of anal-
ysis results to some extent, the determination of an integral radius in the nonlocal models or a gradient parameter in the
gradient models is an additional task. It is indeed pointed out that the physical meanings of these parameters are ambiguous
[18,19].
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 51

Several studies have also been made on introducing fracture mechanics model, which is called cohesive crack or cohesive
zone model, into the finite element analysis for crack propagation. Hillerborg et al. [20] have proposed the energy balance
approach as one of the cohesive model based on fracture mechanics for concrete in terms of fracture energy. In the cohesive
model, cohesive tractions are assumed to act between crack surfaces to represent a fracture process zone in concrete. To sim-
ulate arbitrary propagating cracks in concrete using finite elements, the cohesive model has recently been incorporated with
the PU-based FEM; see for example the X-FEM [21], PUFEM [22], FCM [23] and other related methods [24,25]. However, little
attention has been paid to the combination of the cohesive model with the methods which replace the cracking by the degra-
dation of stiffness represented by a continuum damage model, because the cohesive model is generally applied on the crack
surfaces where cohesive tractions are transferred.
In this context, the purpose of this study is to propose an isotropic damage model based on fracture mechanics, which
represents strain-softening behavior associated with cracking, without introducing the above-mentioned conventional non-
local models. In order to verify the validity and effectiveness of the proposed model, we solve representative example prob-
lems of crack propagation in concrete. Section 2 starts with the formulation of the 1D damage model, which incorporates the
well-known fracture mechanics model based on energy balance [20]. Then, we extend the model to multi-dimensional prob-
lems by using the equivalent strain of the modified von-Mises model [27]. In Section 3, verification is conducted to demon-
strate the promise and capability of the proposed model for evaluating energy balance in deformation/fracture processes
with high accuracy and without depending on element type and size. A benchmark problem involving mixed-mode fracture
is also solved in Section 4 to demonstrate the ability of the proposed model in reproducing the corresponding experimental
result.

2. Formulation of damage model based on fracture mechanics

The formulation of the proposed isotropic damage model is presented in this section. After incorporating the cohesive
crack model into the 1D damage model within the framework of fracture mechanics, we extend the derived 1D model to
multi-dimensional problems using the equivalent strain of the modified von-Mises model [27].

2.1. Formulation in 1D problems

2.1.1. Stress–strain relation for damage model

In the standard 1D damage model, the Hooke’s law can be expressed as
r ¼ ð1  DÞEe ð1Þ
where r; e and E are the stress, strain and Young’s modulus, respectively. Scalar variable D is called the damage variable that
represents the degree of mechanical degradation and takes 0 6 D 6 1. Here, zero means no damage, while 1 corresponds to
complete fracture.

2.1.2. Relationship between cohesive force and crack opening displacement

Using an exponential function, we assume the following relationship in a 1D setting:

f ¼ AeBw ð2Þ
where f is the cohesive force representing stress transfer on a fracture surface, w is the crack opening displacement (COD) of
the fracture surface, and A and B are unknown material parameters.
Denoting the uniaxial tensile strength by f t , we recognize f equals f t when w ¼ 0. Then, the parameter A can be expressed
as

f t ¼ Ae0 ! A ¼ f t ð3Þ
Also, the fracture energy, denoted by Gf , is defined as the energy per unit area required to form a fracture surface, which
equals the area under the cohesive force-COD curve. According to this definition, the fracture energy must satisfy
Z 1 Z 1
Gf ¼ f dw ¼ A eBw dw ð4Þ
0 0

so that unknown parameter B can be expressed as

A A f
Gf ¼ !B¼ ¼ t ð5Þ
B Gf Gf

Consequently, the cohesive force-COD relationship for 1D problems can be expressed as

 
f
Gt w f
f ¼ f te f ¼ f t exp  t w ð6Þ
Gf
52 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

2.1.3. Incorporation into damage model

The above-presented cohesive force-COD relationship involving the fracture energy is now incorporated into the conven-
tional damage model. To do that, we first define the damage initiation strain, denoted by e0 , which is related to the tensile
strength, as
ft
e0 ¼ ð7Þ
E
Supposing that a single element accommodates a single crack, COD w can be expressed as
w ¼ eh  e0 h ¼ ðe  e0 Þh ð8Þ
where h is the length of an element in which the damage is evaluated, which corresponds to that of finite element in the
finite element analysis. Here, the element length plays a role of converter of the strain measure to the measure of crack open-
ing displacement and vice versa, since a finite element accommodates only the strain as an element variable.
In a damage model, cohesive force f on the fracture surface corresponds to stress r as
f ¼r ð9Þ
Substituting Eqs. (7)–(9) into Eq. (6), we can transform the cohesive force-COD relationship as
 
Ee0 h
r ¼ Ee0 exp  ðe  e0 Þ ð10Þ
Gf
By converting this relationship into the same format as Eq. (1), we have
     
e0 Ee0 h e0 Ee0 h
r¼ exp  ðe  e0 Þ Ee ¼ 1  1  exp  ðe  e0 Þ Ee ¼ ½1  DðeÞEe ð11Þ
e Gf e Gf
Then, the damage variable can be represented as
 
0 E0 h
DðÞ ¼ 1  exp  ð  0 Þ ð12Þ
 Gf
where  P 0 is the maximum strain the material ever experienced.
Comparing the magnitudes of e and , we can judge whether the current condition is loading or unloading such that

if  6 e !  ¼ e ðloadingÞ
ð13Þ
if  > e !  ¼  ðunloadingÞ
Also, the judgment for damaged or undamaged states of the material is given as

if  < 0 ! D ¼ 0 ðundamageÞ
ð14Þ
if  P 0 ! D ¼ DðÞ
Within the framework of conventional continuum damage theory, the functional form of the damage variable is often
given as [10,12–14]
0 
Dref ðÞ ¼ 1  1  a þ aebð Þ 0
ð15Þ

where a and b are the material parameters determined with reference to experimental results. Assuming a ¼ 1, we have
0 bð Þ
Dref ðÞ ¼ 1  e 0
ð16Þ

Compared with Eq. (12), it is found that b is a parameter representing material toughness, as
E0 h
b¼ ð17Þ
Gf
which involves the fracture energy Gf and the element length h resulting form Eq. (8). It is, therefore, recognized that the
proposed model enables us to identify the parameters in Eq. (15) or Eq. (16) with those in fracture mechanics for concrete.

2.2. Extension to multi-dimensional problems

2.2.1. Equivalent strain

In an isotropic damage model for multi-dimensional problems, the damage nucleation and evolution law generally hinges
on the definition of an equivalent strain, which can be used in place of the unidirectional strain e in the 1D model presented
above. Several definitions of equivalent strains have been suggested to describe the fractures of quasi-brittle materials; see
for example, [26,27]. Among them, we employ the modified von-Mises model proposed by de Vree et al. [27], which has been
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 53

proven in a variety of studies on isotropic damage of concrete [10–14]. The equivalent strain eeq in the employed model is
defined as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 2
k1 1 k1 12k
eeq ¼ I1 þ I1 þ J2 ð18Þ
2kð1  2mÞ 2k 1  2m ð1 þ mÞ2
where m and k represent the Poisson’s ratio and the ratio of tensile and compressive strength, respectively. In the case of con-
crete, for example, k P 10 is suggested in general. Also, I1 and J 2 are the first invariant of strain tensor e and the second
invariant of deviatoric strain tensor e ¼ e  ð1=3Þtr e, which are respectively defined as
I1 ¼ tr e ¼ ekk ð19Þ
1 1
J 2 ¼ e : e ¼ ekl ekl ð20Þ
2 2
Fig. 1 shows some profiles of the equivalent strain on a 2D principal strain space with different values of tensile-
compressive strength ratio k. The well-known characteristics of quasi-brittle materials like concrete, which shows low
strength in tension and high strength in compression, is well demonstrated by k.

2.2.2. Damage model using equivalent strain

The direct extension of the Hooke’s law in the 1D damage model, Eq. (1), to multi-dimensional problems can be given as
r ¼ ð1  DÞc : e ð21Þ
where r and c are the Cauchy stress tensor and elastic modulus tensor, respectively. Since the damage variable here takes a
scalar value, implying the lack of directionality, the model is commonly referred to as isotropic damage model.
Denoting the maximum equivalent strain in the deformation history by j P 0, we recognize that the following functional
form of the damage variable can be used for the isotropic damage model:
 
j0 Ej0 he
DðjÞ ¼ 1  exp  ðj  j0 Þ ð22Þ
j Gf
j0
¼ 1  ebðjj0 Þ ð23Þ
j
which is almost the same as that for the 1D model, given in Eq. (12), except that the representative element length he is
adjusted for multi-dimensional problems as follows:
pffiffiffiffiffiffiffiffi
Triangular element : he ¼ 2Ae ð24Þ
1=3
Tetrahedral element : he ¼ ð12V e Þ ð25Þ
pffiffiffiffiffi
Quadrilateral element : he ¼ Ae ð26Þ
Hexahedral element : he ¼ V 1=3
e ð27Þ

0
Principal strain 2

k=1
ν=0

k=1
ν = 0.2
k=2
ν=0

0
Principal strain 1

Fig. 1. Contour plot of modified von-Mises equivalent strain.

54 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

Here, Ae and V e are respectively the area of 2D elements and the volume of 3D elements. The multiplier 2 in Eq. (24) for tri-
angular elements arises from the assumption that two triangles form a quadrilateral and the multiplier 12 in Eq. (25) for
tetrahedral elements follows the same idea. Since a scalar parameter he is used as a representative element length to eval-
uate the damage of the element, the effect of direction of mesh division cannot be taken into account. Eqs. (13) and (14)
apply for the multi-dimensional problems to judge whether loading or unloading, and whether damaged or undamaged,
respectively.

2.2.3. Tangent modulus tensor – material Jacobian

When j < j0 (before damage initiation or in elastic range), D ¼ 0, and the material Jacobian C is expressed as
@r
C¼ ¼c ð28Þ
@e
When j P j0 and j 6 eeq (with damage and loading), D ¼ DðjÞ and j ¼ eeq . Then, stress tensor r and material Jacobian C
are given as
r ¼ ð1  DðjÞÞc : e ð29Þ
@DðjÞ
C ¼ ð1  DðjÞÞc  ðc : eÞ  ð30Þ
@e
According to the chain rule of differentiation, @DðjÞ=@ e can be expressed as
@DðjÞ @DðjÞ @ j @ eeq
¼ ð31Þ
@e @ j @ eeq @ e
where the derivatives are evaluated as follows:
 
@DðjÞ j0 1
¼ þ b ebðjj0 Þ ð32Þ
@j j j
@j
¼1 ð33Þ
@ eeq
 
@ eeq a1 @I1 1 12 @I1 @J
¼ þ a2 I2 þ a2 J 2 2a21 I1 þ a2 2 ð34Þ
@e 2k @ e 4k 1 1 @e @e
Here, we have defined a1 and a2 as
k1 12k
a1 ¼ ; a2 ¼ ð35Þ
1  2m ð1 þ mÞ2
When j P j0 and j P eeq (with damage and unloading), the effect of damage is present, but the damage never propa-
gates. Then, j is not updated as j ¼ j so that Eq. (31) yields
@DðjÞ @DðjÞ @ eeq
¼ 0 ¼0 ð36Þ
@e @j @e
Consequently, the material Jacobian C becomes
C ¼ ð1  DðjÞÞc ð37Þ
In numerical analyses with the proposed model, the implicit solution procedure based on the standard Newton–Raphson
(NR) method is employed. If the convergent solution is not obtained due to the numerical instability especially in the soft-
ening regime, we return to the initial iteration step of the incremental process and apply the modified NR method instead.

3. Performance demonstrations

Several numerical examples are presented to verify the capability of the model for evaluating the fracture process with
high accuracy and to demonstrate its performance for providing mesh-independent solutions.

3.1. Fracture under uniaxial tension

A numerical examination is conducted to verify the basic characteristics of the proposed damage model for uniaxial ten-
sile fractures. Fig. 2 shows the analysis targets along with meshes, which are prepared so as to have damage only in the ele-
ments aligned in their center. Four-types of 2D and 3D finite element (FE) meshes of different linear elements are prepared to
examine the element lengths given in Eqs. (24)–(27); they are standard triangular (T3), quadrilateral (Q4), tetrahedral (T4)
and hexahedral elements (H8). Here, the term ‘‘standard” to explain elements implies throughout this study that these
elements are not endowed with special techniques such as reduced integrations and introduction of incompatible modes.
Material parameters are provided in the figure, and the Poisson’s ratio is set at zero to prevent the Poisson effect.
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 55

110 mm

Displacement: 0.1 mm
40 mm
110 mm

Displacement: 0.1 mm
40 mm

Thickness: 20 mm

E ν k κ0 Gf

20000 MPa 0.0 10 0.0001 0.05 N/mm

Fig. 2. Uniaxial tensile fracture problem.

: Q4
: T3
Load / Area (MPa)

0
0 0.0005
Displacement / Length
(a) Load-displacement curves in 2-D simulation

2
: H8
: T4
Load / Area (MPa)

0
0 0.0005
Displacement / Length
(b) Load-displacement curves in 3-D simulation

Fig. 3. Load versus displacement in uniaxial problem.

56 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

Displacement: 1.5 mm
Thickness: 100 mm

100 mm
Ligament C

800 mm

Model C40 C = 40 (mm)

Model C50 C = 50 (mm)

Model C60 C = 60 (mm)

E ν k κ0 Gf

30000 MPa 0.0 10 0.0001 0.10 N/mm

Fig. 4. Three-point bending test on beam with a differently-sized single-edge notch.

: C40
: C50
1
: C60
Load (kN)

0
0 1
Displacement (mm)

Fig. 5. Load versus displacement in 3-point bending problem.

Fig. 3 shows the load–displacement curves obtained by the 2D and 3D simulations with different FE meshes. In the figure,
the ordinate and abscissa represent the apparent stress and strain, respectively. As can be seen from these curves, the results
obtained are almost the same for all the element types. Although the results of the tetrahedral elements are slightly different
from that of the hexahedral elements, a source of error seems to be the assumption about element length that 12 tetrahe-
drons form a hexahedron; see Eq. (25).

3.2. Fracture in beams with a notch under three-point bending

Fracture energy, which is one of the representative fracture mechanics parameters for concrete, is defined as the energy
per unit area required to form a fracture surface or the energy dissipated when the fracture surface of a unit area is formed. In
this context, Li et al. [28] have proposed a method of evaluating the fracture energy of concrete by using the J-integral. In the
method, three-point bending tests are performed on beam specimens with different notch lengths, and then the differences
in dissipated energy and in notch length enable us to estimate the fracture energy. Here, following this idea, we conduct
numerical examination to evaluate the fracture energy and discuss the energy characteristics of the proposed damage model.
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 57

3.2.1. Analysis target and conditions

The analysis target is the beam with a notch subjected to three-point bending as shown in Fig. 4, in accordance with
the RILEM method [29]. We prepare three FE models with the same span, height and thickness, but different notch lengths.
The finite elements used here are the standard bilinear quadrilateral elements. Setting the material parameters as provided
in the figure, we conduct crack propagation analyses under displacement control.

3.2.2. Analysis results and discussion

Fig. 5 shows the load–displacement curves obtained by the numerical analyses with the proposed model. By calculating
the area under these curves, we obtain the total energy dissipated in the fracture process in each model, which corresponds
to the energy necessary to form the total area of fracture surface in the three-point bending test. Here, the difference in total
dissipated energy between two separate models corresponds to the difference in total area of fracture surface. Therefore, the
fracture energy, namely the energy dissipation per unit area, can be evaluated by dividing this difference in total dissipated
energy by the difference in area of fracture surface.
For example, denoting the difference in dissipated energy between Model C50 and C40 by DU 5040 and the difference in
height minus notch length times thickness by DA ð¼ 10  100 mm2), we can calculate the apparent fracture energy as
GF ¼ DU=DA, which can be compared with the fracture energy Gf set as a material parameter in our numerical analyses.
The apparent fracture energy GF5040 obtained from Model C50 and C40, and the apparent fracture energy GF6050 obtained
from Model C60 and C50, are respectively calculated as follows:

4:57500  102  3:5802  102

G5040 ¼ ¼ 0:09948     0:1 N=mm
F
ð50  40Þ  100
5:57980  102  4:57500  102
G6050 ¼ ¼ 0:10048     0:1 N=mm
F
ð60  50Þ  100

Here, the apparent fracture energies GF5040 and GF6050 , estimated from the numerical analyses, reproduce the material
parameter Gf ¼ 0:1 N/mm with satisfactory accuracy. This result confirms that the simulated crack behavior with strain-
softening meets the definition of fracture energy. Therefore, the proposed damage model enables us to perform crack prop-
agation analyses in consistency with fracture mechanics for concrete.

3.3. Flexural fracture in arch-shaped structure

The proposed damage model contains the ratio of tensile and compressive strength denoted by k as shown in Eq. (18),
which enables us to simulate the fracture behavior of concrete such that the compressive strength is higher than the tensile
one. We here examine an arch-shaped structure subjected to bending moment for the effect of the ratio in the damage model
on the fracture behavior.

mm 100 mm
40

m
4m
nt: 0.
e
cem
ispla
D
200 mm

z x

E ν k κ0 Gf

20000 MPa 0.2 1, 10 0.0001 0.1 N/mm

Fig. 6. Fluxural fracture problem of arch-shaped model.

58 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

100/400 400/400
(a) Damage distribution (k = 1)

100/400 400/400
(b) Damage distribution (k = 10)

0.0 1.0
Fig. 7. Propagation of damage in arch-shaped model.

120
Load (N)

80

40
:k=1
: k = 10

0
0 0.2 0.4
Displacement (mm)

Fig. 8. Load versus displacement in arch-shaped model.

 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 59

Displacement: 0.05 mm

Model H1

100 mm
Displacement: 0.1 mm Displacement: 0.1 mm

200 mm 200 mm
Model H2 Model H3

E ν k κ0 Gf

20000 MPa 0.1 10 0.0001 0.04 N/mm

Fig. 9. Tensile fracture problem of differently-sized plate with a circular hole.

: H1
: H2
Load / Area (MPa)

1 : H3

0
0 0.0002 0.0004
Displacement / length

Fig. 10. Load versus displacement in circular hole problem.

The analysis target is the arch-shaped structure under the boundary condition giving rise to bending moment as shown in
Fig. 6. The standard linear tetrahedral elements are used to prepare the FE model. The material prameters set in this example
are shown in the same figure. Crack propagation analyses are carried out under displacement control by varying the ratio of
tensile and compressive strength.
Fig. 7 shows the distribution of damage for the two analysis cases. In case of k ¼ 1, damages start from both the tensile
and compressive sides of the arch-shaped structure. This is because the compressive strength is the same as the tensile one.
In contrast, as can be seen from the case of k ¼ 10, the damage arises only from the tensile side due to the condition that the
60 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

Loading step: 20/150 Loading step: 150/150

(a) Model H1

Loading step: 20/150 Loading step: 150/150

(b) Model H2

Loading step: 20/150 Loading step: 150/150

(c) Model H3
0.0 1.0

Fig. 11. Propagation of damage in circular hole problem.

compressive strength is ten times higher than the tensile one. Therefore, setting the ratio of strength to a large value makes it
possible to simulate realistic crack propagations as observed in concrete.
Fig. 8 shows the load–displacement curves for the two analysis cases. As can be seen from the figure, the responses are
different. The results show that the ratio of strength has influences on not only crack propagations but also apparent load–
displacement responses.

3.4. Capability for mesh independency

It is known that damage models without any special treatment suffers from the strong mesh dependency in simulating
fractures involving strain-softening behavior [15,27]. In this section, using the proposed model, we perform a case study of
fracture simulations to demonstrate its capability for mesh-independency.
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 61

The analysis target is a plate with a circular hole. For the case study, we prepare three kinds of FE meshes as shown in
Fig. 9. Model H1 is set for reference. Also, Model H2 is prepared to be double the mesh size and the model size of H1,
and H3 has the same mesh size as H1 with double the model size of H1. Standard linear triangular elements are randomly
arranged to form the meshes. The material parameters used in the case study are provided in the figure, and the plane strain
conditions are assumed in the finite element analyses.
Fig. 10 shows the load–displacement curves for the three analysis cases. The ordinate indicates the apparent stress that is
defined as the load divided by the loading area, while the abscissa represents the apparent strain that is the displacement
divided by the height of the model. As can be seen from the results, the responses of Model H2 and Model H3, which are
the same in model size but different in mesh size, are almost the same. These results thus confirmed that the fracture anal-
ysis with the proposed model do not depend on mesh size. On the other hand, comparing the results of H1 and H2 (or H3),
which are different in model size, we can see that Model H2 (or H3) with a larger model size exhibits rather brittle response.
This type of dependency of the strain-softening responses on the model size – a well-known characteristic of fracture behav-
ior – has been reproduced by the application of the proposed damage model.
Fig. 11 shows the damage evolution for all the cases. Here, the values D ¼ 0 and D ¼ 1 are colored in white and black,
respectively, and the gray scale is the intermediate value of D. Again, it is found that the locations and distribution properties
of the damage variable do not depend on the mesh size, when the results of Model H2 and H3 are compared. From the results
and discussions here, we can confirm that failure simulations with the proposed model, which satisfies the energy balance in
terms of the fracture energy, reflect the effect of model size without dependency on the mesh size.

4. A benchmark problem for mixed-mode fracture

Taking a benchmark problem for mixed-mode fracture as the final numerical example, we carry out the crack propagation
analysis with the proposed damage model, to verify the reproducibility of the experimental result.

4.1. Nooru-Mohamed test

The experiments conducted by Nooru-Mohamed [30], in which tensile (Mode I) and shear (Mode II) loads are applied to a
mortar specimen, are often referred as benchmark problems for mixed-mode fracture in concrete or other quasi-brittle
materials, and their results have widely be utilized to validate the numerical results obtained by various techniques for frac-
ture simulations [11,31–34].
Fig. 12 shows a schematic of the rectangular parallelepiped specimen along with the boundary conditions and the rep-
resentative crack paths obtained in one of the experiments. Here, the solid and dotted curves are the crack paths observed
on the front and reverse sides of the specimen. Two notches are introduced to the specimen, which is sometimes called DENS
(Double Edge Notched Specimen). Stiff plates (T, B, L, R in the figure) are attached to the four sides of the specimen to apply
Displacement: 0.1 mm
(Mode I)

T
Concrete
Load: 10 kN

(Mortar)
(Mode II)

97.5 mm

L
5 mm

Load: 10 kN
(Mode II)
97.5 mm

R
Stiff platen 200 mm
y
B
x
Thickness: 50 mm

E ν k κ0 Gf

30000 MPa 0.2 10 0.0001 0.11 N/mm

Fig. 12. Nooru-Mohamed problem.

62 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

4,112 nodes, 4,024 elements 5119 nodes, 10,112 element

(a) 2D mesh with quadrilateral elements (b) 2D mesh with triangular elements

65,792 nodes, 60,360 elements

Front view 3D view

(c) 3D mesh with hexahedral elements

112,413 nodes, 625,252 elements

Front view 3D view

(d) 3D mesh with tetrahedral elements
Fig. 13. Finite element mesh for Nooru-Mohamed problem.

the loading. With plates L and R being fixed and kept upright, the constant horizontal pressure is applied to both of the plates
from the beginning to the end, and the displacement is incrementally given to plate T in the vertical direction. Although the
experiments were conducted with different values of horizontal pressure, we here employ only the case of applied horizontal
load of 10 kN.

4.2. Analysis conditions

As shown in Fig. 13, the FE meshes with four different element types in 2D and 3D are used for fracture simulations for the
Nooru-Mohamed problem. One of the 2D meshes is generated with standard four-node quadrilateral elements and the other
is with standard three-node triangular elements. The 3D meshes are generated with standard eight-node hexahedral and
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 63

Loading step: 30/200 Loading step: 200/200

(a) 2D mesh with quadrilateral elements

Loading step: 30/200 Loading step: 200/200

(b) 2D mesh with triangular elements
0.0 1.0

Fig. 14. Propagation of damage in Nooru-Mohamed problem: 2D results.

four-node tetrahedral elements. Bilateral and diphycercal symmetries are kept in these meshes. The stiff plates in each mesh
are also generated with the same type of elements with that of the mortar, and are connected to the mortar domain. The
material parameters used in the numerical analyses are provided in the figure, and the Young’s modulus of the stiff plates
is set to 100 times that of the mortar specimen. 2D analyses are carried out under the plane strain conditions.
With the y-component of the displacement on the lower surface of plate B being set at zero, constant horizontal loads of
10 kN are applied to the left-end of plate L and the right-end of plate R from the beginning to the end, while the forced dis-
placement only in the y-direction is incrementally applied on the upper surface of plate T. Expecting mirror-symmetric
deformations, we impose no constraints of the x-component of displacement on all the plates’ edges. Although the
detachment of the mortar part from the stiff plates might be possible in actual experiments, we do not consider such imper-
fect situations in this study.

4.3. Analysis results and discussion

Figs. 14 and 15 show the distributions of damage variables at two stages of evolutions, which approximately represent
crack paths, along with the experimental ones. Here, the meanings of white and black colors are the same as before. With
reference to Fig. 12, the numerical analyses with the proposed damage model provided very similar results to the experi-
mentally obtained crack paths. It should be emphasized that the fracture nucleation and propagation in the numerical anal-
yses are determined only by the tensile strength in terms of the equivalent strain of the modified von-Mises model, which
involves the elastic properties and the ratio of tensile and compressive strength. In other words, the numerical results with
the proposed damage model reproduced the mixed-mode fracture characteristics similar to the experimental result, even
though Mode I and Mode II fractures are not distinguished in the material model.
Fig. 16 shows the obtained load–displacement curves along with the two experimental results provided in the [30]. For
reference, the result of the X-FEM analysis with the same set of material parameters as the present analysis, which is
reported in [33], is also provided in the figure. As can be seen from these curves, the analysis results with the proposed model
are comparable with the experimental ones. More specifically, the softening responses of the numerical results are con-
formable with the experimental ones. However, when the numerical results are mutually compared, they are moderately
different especially in the softening behavior after the peak loads.
64 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

Loading step: 30/200 Loading step: 150/200

(a) 3D mesh with hexahedral elements

Loading step: 30/200 Loading step: 150/200

(b) 3D mesh with tetrahedral elements

Fig. 15. Propagation of damage in Nooru-Mohamed problem: 3D results.

When the 2D results with quadrilateral and triangular elements are compared, the triangular mesh provides more brittle
behavior than the structured mesh with quadrilateral elements. Also, the results of the latter model, namely those of the
structured mesh with quadrilateral elements, exhibit zigzag crack paths in an oblique direction, as can be observed in
Fig. 14. Furthermore, the crack paths obtained with triangular elements seem to be narrower than those with quadrilateral
elements. The difference in softening behavior or brittleness between the results with triangular and quadrilateral elements
 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66 65

20
: Quadrilateral mesh
: Triangular mesh
: X-FEM (reference)

Load (kN)
10
Exp. results

0
0 0.1
Displacement (mm)
(a) 2D results

20
: Hexahedral mesh
: Tetrahedral mesh
Load (kN)

10

0
0 0.1
Displacement (mm)
(b) 3D results

Fig. 16. Load versus displacement in Nooru-Mohamed problem.

implies that the softening behavior characterized with the proposed model may be slightly affected by element types and
mesh topologies especially for non-linear crack paths. As can be seen from Fig. 16, the same tendency of softening behavior
is observed for the 3D results with hexahedral and tetrahedral elements, though the difference in crack paths is not signif-
icant in comparison with the 2D results.
The final remark is addressed on the effect of Poisson’s ratio. Since the present study employs the equivalent strain based
on the modified von-Mises model for the fracture criteria, the value of the Poisson’s ratio affects both shear and volumetric
deformations especially for 2D analyses under the plane strain condition. Although the results in this example problem are
obtained with Poisson’s ratio of 0.2, we have confirmed that the smaller the value of Poisson’s ratio, the larger the peak
strength in 2D analyses. Moreover, we have also confirmed that the load–displacement curve coincides with that of the
X-FEM provided in Fig. 16 as a reference solution, if the Poisson’s ratio is set at 0.0 in 2D analysis under the plane strain
condition.

5. Conclusion

To avoid mesh-size dependency, most of the previous studies using damage models employ nonlocal models such as the
integral average model and the gradient model, which require parameters whose physical interpretation is ambiguous. The
nonlocal parameters in such damage models are associated with the profile of material’s distribution, but not with material
properties in fracture mechanics, to define the damage evolution law.
In this paper, without using any conventional nonlocal model of such as those with integral average or gradient, we pro-
posed an isotropic damage model to characterize the strain-softening behavior within the framework of fracture mechanics
for concrete. After presenting the formulation in 1D problems by incorporating the cohesive force-COD relationship of con-
crete into the damage model in terms of the fracture energy, we extended it to multi-dimensional problems using the equiv-
alent strain based on the modified von-Mises model. A series of crack propagation analyses has been conducted to verify the
validity of the proposed model and to demonstrate its performance and capability in simulating mesh-independent fracture
responses involving the strain-softening behavior and in reproducing experimentally reported crack paths under the mixed-
mode loading.
66 M. Kurumatani et al. / Engineering Fracture Mechanics 155 (2016) 49–66

While our focus has been placed on the fracture processes mainly caused by locally tensile cracking, the applicability of
the proposed model to fracture phenomena under compressive loading has to be examined in the near future. Also, the
investigation of the effects of heterogeneities on the fracture behavior of concrete is an important subject of study.

Acknowledgements

This work was supported by the Ministry of Education, Culture, Sports, Science and Technology (MEXT), KAKENHI Grant
Nos. 24760357 and 22360176. These supports are gratefully acknowledged.

References

[1] Jirásek M. Comparative study on finite elements with embedded discontinuities. Comput Methods Appl Mech Engng 2000;188:307–30.
[2] Moës N, Dolbow J, Belytschko T. A finite element method for crack growth without remeshing. Int J Numer Meth Engng 1999;46:131–50.
[3] Terada K, Asai M, Yamagishi M. Finite cover method for linear and nonlinear analyses of heterogeneous solids. Int J Numer Meth Engng
2003;58:1321–46.
[4] Wells GN, Sluys LJ. Application of embedded discontinuities for softening solids. Engng Fract Mech 2000;65:263–81.
[5] Kurumatani M, Terada K. Finite cover method with multi-cover-layers for the analysis of evolving discontinuities in heterogeneous media. Int J Numer
Meth Engng 2009;79:1–24.
[6] Bažant ZP, Oh BH. Crack band theory for fracture of concrete. Mater Struct 1983;16:155–77.
[7] Jirásek M, Zimmermannl T. Analysis of rotating crack model. J Engng Mech 1998;124:842–51.
[8] Jirásek M, Zimmermann T. Rotating crack model with transition to scalar damage. J Engng Mech 1998;124:277–84.
[9] Bažant ZP, Ožbolt J. Nonlocal microplane model for fracture, damage and size effect in structures. J Engng Mech 1990;116:2485–505.
[10] Geers MGD, de Borst R, Peerlings RHJ. Damage and crack modeling in single-edge and double-edge notched concrete beams. Engng Fract Mech
2000;65:247–61.
[11] Jirásek M, Grassl P. Evaluation of directional mesh bias in concrete fracture simulations using continuum damage models. Engng Fract Mech
2008;75:1921–43.
[12] Peerlings RHJ, de Borst R, Brekelmans WAM, Geers MGD. Gradient-enhanced damage modelling of concrete fracture. Mech Cohes-Frict Mater
1998;3:323–42.
[13] Rodríguez-Ferran A, Bennett T, Askes H, Tamayo-Mas E. A general framework for softening regularisation based on gradient elasticity. Int J Solids Struct
2011;48:1382–94.
[14] Giry C, Dufour F, Mazars J. Stress-based nonlocal damage model. Int J Solids Struct 2011;48:3431–43.
[15] Pijaudier-Cabot G, Bažant ZP. Nonlocal damage theory. J Engng Mech 1987;113:1512–33.
[16] Bažant ZP, Pijaudier-Cabot G. Nonlocal damage, localization instability and convergence. J Appl Mech 1988;55:287–93.
[17] Peerlings RHJ, de Borst R, Brekelmans WAM, de Vree JHP. Gradient enhanced damage for quasi-brittle materials. Int J Numer Meth Engng
1996;39:3391–403.
[18] Bažant ZP, Pijaudier-Cabot G. Measurement of characteristic length of nonlocal continuum. J Engng Mech 1989;115:755–67.
[19] Kuhl E, Ramm E, de Borst R. An anisotropic gradient damage model for quasi-brittle materials. Comput Methods Appl Mech Engng 2000;183:87–103.
[20] Hillerborg A, Modéer M, Petersson PE. Analysis of crack formation and crack growth in concrete by means of fracture mechanics and finite elements.
Cem Concr Res 1976;6:773–82.
[21] Moës N, Belytschko T. Extended finite element method for cohesive crack growth. Engng Fract Mech 2002;69:813–33.
[22] Gasser TC, Holzapfel GA. Modeling 3D crack propagation in unreinforced concrete using PUFEM. Comput Methods Appl Mech Engng
2005;194:2859–96.
[23] Terada K, Ishii T, Kyoya T, Kishino Y. Finite cover method for progressive failure with cohesive zone fracture in heterogeneous solids and structures.
Comput Mech 2007;39:191–210.
[24] Wells GN, Sluys LJ. A new method for modelling cohesive cracks using finite elements. Int J Numer Meth Engng 2001;50:2667–82.
[25] Mergheim J, Kuhl E, Steinmann P. A finite element method for the computational modelling of cohesive cracks. Int J Numer Meth Engng
2005;63:276–89.
[26] Mazars J. A description of micro- and macroscale damage of concrete structures. Engng Fract Mech 1986;25:729–37.
[27] de Vree JHP, Brekelmans WAM, van Gils MAJ. Comparison of nonlocal approaches in continuum damage mechanics. Comput Struct 1995;55:581–8.
[28] Li VC, Chan CM, Leung CKY. Experimental determination of the tension-softening relations for cementitious composites. Cem Concr Res
1987;17:441–52.
[29] RILEM Draft Recommendation (50-FMC). Determination of the fracture energy of mortar and concrete by means of three-point bend tests on notched
beams. Mater Struct 1985;18:285–90.
[30] Nooru-Mohamed MB. Mixed-mode fracture of concrete: an experimental approach. Ph.D. Thesis, Technische Universiteit Delft; 1992.
[31] di Prisco M, Ferrara L, Meftah F, Pamin J, de Borst R, Mazars J, et al. Mixed mode fracture in plain and reinforced concrete: some results on benchmark
tests. Int J Fract 2000;103:127–48.
[32] Cervera M, Chiumenti M. Smeared crack approach: back to the original track. Int J Numer Anal Meth Geomech 2006;30:1173–99.
[33] Dumstorff P, Meschke G. Crack propagation criteria in the framework of X-FEM-based structural analyses. Int J Numer Anal Meth Geomech
2007;31:239–59.
[34] Al-Rub RKA, Kim S-M. Computational applications of a coupled plasticity-damage constitutive model for simulating plain concrete fracture. Engng
Fract Mech 2010;77:1577–603.